FAST TRACK

GENE EXPRESSION PROFILES IN PRIMARY OVARIAN SEROUS PAPILLARYTUMORS AND NORMAL OVARIAN EPITHELIUM: IDENTIFICATION OFCANDIDATE MOLECULAR MARKERS FOR OVARIAN CANCER DIAGNOSISAND THERAPYAlessandro D. SANTIN

1,* Fenghuang ZHAN2, Stefania BELLONE

1, Michela PALMIERI1, Stefania CANE

1, Eliana BIGNOTTI1,

Simone ANFOSSI1, Murat GOKDEN

3, Donna DUNN1, Juan J. ROMAN

1, Timothy J. O’BRIEN1, Erming TIAN

2, Martin J. CANNON4,

John SHAUGHNESSY, JR.2 and Sergio PECORELLI5

1Division of Gynecologic Oncology, Department of Obstetrics and Gynecology, University of Arkansas for Medical Sciences,Little Rock, AR, USA2Department of Medicine, Myeloma Institute for Research and Therapy, University of Arkansas for Medical Sciences,Little Rock, AR, USA3Department of Pathology, University of Arkansas for Medical Sciences, Little Rock, AR, USA4Department of Microbiology and Immunology, University of Arkansas, Little Rock, AR, USA5Division of Gynecologic Oncology, University of Brescia, Brescia, Italy

With the goal of identifying genes with a differential pat-tern of expression between ovarian serous papillary carcino-mas (OSPCs) and normal ovarian (NOVA) epithelium andusing this knowledge for the development of novel diagnosticand therapeutic markers for ovarian cancer, we used oligo-nucleotide microarrays with probe sets complementary to12,533 genes to analyze the gene expression profiles of 10primary OSPC cell lines, 2 established OSPC cell lines (UCI-101, UCI-107) and 5 primary NOVA epithelial cultures. Un-supervised analysis of gene expression data identified 129 and170 genes that exhibited >5-fold upregulation and downregu-lation, respectively, in primary OSPC compared to NOVA.Genes overexpressed in established OSPC cell lines had littlecorrelation with those overexpressed in primary OSPC, high-lighting the divergence of gene expression that occurs as aresult of long-term in vitro growth. Hierarchical clustering ofthe expression data readily distinguished normal tissue fromprimary OSPC. Laminin, claudin 3, claudin 4, tumor-associatedcalcium signal transducers 1 and 2 (TROP-1/Ep-CAM, TROP-2),ladinin 1, S100A2, SERPIN2 (PAI-2), CD24, lipocalin 2, osteopon-tin, kallikrein 6 (protease M), kallikrein 10, matriptase (TADG-15) and stratifin were among the most highly overexpressedgenes in OSPC compared to NOVA. Downregulated genes inOSPC included transforming growth factor-� receptor III, plate-let-derived growth factor receptor �, SEMACAP3, ras homologgene family member I (ARHI), thrombospondin 2 and disabled-2/differentially expressed in ovarian carcinoma 2 (Dab2/DOC2).Differential expression of some of these genes, including clau-din 3, claudin 4, TROP-1 and CD24, was validated by quanti-tative RT-PCR and flow cytometry on primary OSPC andNOVA. Immunohistochemical staining of formalin-fixed,paraffin-embedded tumor specimens from which primaryOSPC cultures were derived further confirmed differentialexpression of CD24 and TROP-1/Ep-CAM markers on OSPCvs. NOVA. These results, obtained with highly purified pri-mary cultures of ovarian cancer, highlight important molec-ular features of OSPC and may provide a foundation for thedevelopment of new type-specific therapies against this dis-ease.© 2004 Wiley-Liss, Inc.

Key words: serous papillary ovarian cancer; gene expression profil-ing

Ovarian carcinoma remains the cancer with the highest mortal-ity rate among gynecologic malignancies, with 25,400 new cancercases estimated in 2003 in the United States alone.1 OSPC repre-sents the most common histologic type of ovarian carcinoma.2Because of the insidious onset of the disease and the lack ofreliable screening tests, two-thirds of patients have advanced dis-ease when diagnosed; and although many patients with dissemi-nated tumors respond initially to standard combinations of surgical

and cytotoxic therapy, nearly 90% will develop recurrence andinevitably succumb to their disease.2 Understanding the molecularbasis of OSPC may significantly refine the diagnosis and manage-ment of these tumors and may eventually lead to the developmentof novel, more specific and more effective treatment modalities.

cDNA microarray technology has been used to identify genesinvolved in ovarian carcinogenesis.3–11 Gene expression finger-prints representing large numbers of genes may allow precise andaccurate grouping of human tumors and may identify patients whoare unlikely to be cured by conventional therapy. Consistent withthis view, evidence has been provided to support the notion thatpoor-prognosis B-cell lymphomas and biologically aggressivebreast and ovarian carcinomas can be readily distinguished bygene expression profiles.4,12,13 In addition, large-scale gene expres-sion analysis has the potential to identify a number of differentiallyexpressed genes in OSPC compared to NOVA epithelial cells andmay therefore lay the groundwork for future studies of some ofthese markers for clinical utility in the diagnosis and treatment ofOSPC.

Expression profiling studies involve the comparison of OSPCcells with their normal counterpart. However, selection of NOVAcontrol cells may strongly influence the determination of differen-tially expressed genes in ovarian cancer expression profiling stud-ies.14 In this regard, because ovarian epithelial cells constitute only

Abbreviations: CPE, Clostridium perfringens enterotoxin; Ct, cyclethreshold; Ep-CAM, epithelial cell adhesion molecule; FDR, false discov-ery rate; FIGO, International Federation of Gynecology and Obstetrics;MAb, monoclonal antibody; MFI, mean fluorescence intensity; NOVA,normal ovarian; OSPC, ovarian serous papillary carcinoma; PAI, plasmin-ogen activator inhibitor; SAM, significance analysis of microarrays; TNF,tumor necrosis factor; WRS, Wilcoxon’s rank sum.

Grant sponsor: Angelo Nocivelli and Camillo Golgi Foundation.

*Correspondence to: Department of Obstetrics and Gynecology, Uni-versity of Arkansas for Medical Sciences Medical Center, 4301 W.Markham, Little Rock, AR 72205-7199, USA. Fax: �501-686-8091.email: santinalessandrod@uams.edu

Received 5 September 2003; Accepted after revision 20 April 2004

DOI 10.1002/ijc.20408Published online 2 June 2004 in Wiley InterScience (www.interscience.

wiley.com).

Int. J. Cancer: 112, 14–25 (2004)© 2004 Wiley-Liss, Inc.

Publication of the International Union Against Cancer

a small proportion of the total cells found in NOVA samples andbecause ovarian tumor tissue contains significant numbers of con-taminant stromal cells as well as a variety of host-derived immunecells (e.g., monocytes, dendritic cells, lymphocytes), we usedoligonucleotide microarrays with probe sets complementary to12,533 genes to analyze the gene expression profiles in primaryOSPC cultures and NOVA epithelium cell lines. Short-term pri-mary OSPC and NOVA cell cultures, minimizing the risk ofselection bias inherent in any long-term in vitro growth, mayprovide an opportunity to study differential gene expression be-tween relatively pure populations of tumor- and NOVA-derivedepithelial cells.

We report that mRNA fingerprints readily distinguish OSPCfrom NOVA epithelial cells and identify a number of genes highlydifferentially expressed between nontransformed ovarian epitheliaand OSPC. Several of the genes identified are already known to beoverexpressed in ovarian cancer, validating our experimental ap-proach as well as our criteria to determine the genes differentiallyexpressed; but several represent novel findings. Of interest, manyof the genes upregulated in ovarian cancer represent surface orsecreted proteins, such as laminin, claudin 3, claudin 4, tumor-associated calcium signal transducers 1 and 2 (TROP-1/Ep-CAM,TROP-2), ladinin 1, S100A2, SERPIN2 (PAI-2), CD24, lipocalin2, osteopontin, kallikrein 6 (protease M), kallikrein 10, matriptase(TADG-15) and stratifin. Quantitative real-time PCR was used tovalidate differences in gene expression between nontransformedovarian epithelia and OSPC for some of these genes, includingTROP-1/Ep-CAM, CD24, claudin 3 and claudin 4. TROP-1/Ep-CAM and CD24 gene expression was further validated throughflow-cytometric analysis of primary OSPC and NOVA as well asby immunohistochemical analysis of archival OSPC and NOVAspecimens.

MATERIALS AND METHODS

Establishment of OSPC and NOVA primary cell linesA total of 15 primary cell lines (10 OSPC and 5 NOVA) were

established after sterile processing of samples from surgical biop-sies, as previously described for ovarian carcinoma speci-mens.3,6,15 UCI-101 and UCI-107, 2 previously characterizedOSPC cell lines,16,17 were also included. All fresh tumor sampleswere obtained with appropriate consent according to institutionalreview board guidelines. Tumors were staged according to theFIGO operative staging system. Radical tumor debulking, includ-ing total abdominal hysterectomy and omentectomy, was per-formed in all ovarian carcinoma patients, while NOVA tissue wasobtained from consenting donors of similar age undergoing sur-gery for benign pathology. No patient received chemotherapybefore surgery. Patient characteristics are described in Table I.Briefly, normal tissue was obtained by scraping epithelial cellsfrom the ovarian surface and placing them in RPMI-1640 medium(Sigma, St. Louis, MO) containing 10% FBS (Invitrogen, GrandIsland, NY), 200 �/ml penicillin and 200 �g/ml streptomycin.Epithelial explants were then allowed to attach and proliferate.Once the epithelial cells reached confluence, explants were

trypsinized and subcultured for 3 or 4 passages before beingcollected for RNA extraction. Viable tumor tissue was mechani-cally minced in RPMI-1640 to portions no larger than 1–3 mm3

and washed twice with RPMI-1640. Portions of minced tumorwere then placed into 250 ml flasks containing 30 ml of enzymesolution [0.14% collagenase type I (Sigma) and 0.01% DNAse(Sigma, 2,000 KU/mg)] in RPMI-1640 and incubated in a mag-netic stirring apparatus overnight at 4°C. Enzymatically dissoci-ated tumor was then filtered through 150 �m nylon mesh togenerate a single-cell suspension. The resultant cell suspensionwas then washed twice in RPMI-1640 plus 10% FBS. Primary celllines were maintained in RPMI-1640 supplemented with 10%FBS, 200 �/ml penicillin and 200 �g/ml streptomycin at 37°C, 5%CO2 in tissue culture flasks (75–150 cm;2 Corning, Corning, NY).Tumor cells were collected for RNA extraction at 50–80% con-fluence after 2–12 passages in vitro. The epithelial nature andpurity of OSPC and NOVA cultures were verified by immunohis-tochemical staining and flow-cytometric analysis with antibodiesagainst cytokeratin, as previously described.3,15 Only primary cul-tures with at least 90% viability and �99% epithelial cells wereused for total RNA extraction.

RNA purification and microarray hybridization and analysisDetailed protocols for RNA purification, cDNA synthesis,

cRNA preparation and hybridization to the Affymetrix (SantaClara, CA) Human U95Av2 GeneChip microarray were performedaccording to the manufacturer’s protocols, as reported previously.18

Data processingAll data used were derived from Affymetrix 5.0 software.

GeneChip 5.0 output files are given as a signal that represents thedifference between the intensities of the sequence-specific perfect-match probe set and the mismatch probe set or as detection ofpresent, marginal or absent signals as determined by the program’salgorithm. Gene arrays were scaled to an average signal of 1,500and then analyzed independently. Signal calls were transformed bylog base 2, and each sample was normalized to give a mean of 0and variance of 1.

Gene expression data analysisStatistical analyses of the data were performed with the software

package SPSS10.0 (SPSS, Chicago, IL) and the SAM method.19

Genes were selected for analysis based on detection and foldchange. In each comparison, genes having “present” detectioncalls in more than half of the samples in the overexpressed groupwere retained for statistical analysis if they showed a �2-foldchange between groups. Retained genes were subjected to SAM toestablish a FDR, then further filtered via the WRS test at � � 0.05.The FDR obtained from the initial SAM was assumed to charac-terize genes found significant via WRS.

Gene Cluster/TreeviewThe hierarchical clustering of the average-linkage method with

the centered correlation metric was used.20 The dendrogram wasconstructed with a subset of genes from 12,533 probe sets presenton the microarray, whose expression levels varied the most among

TABLE I – CHARACTERISTICS OF THE PATIENTS

Patient Age Race Grade Stage Presence ofascites

Chemotherapyregimen Response to therapy

OSPC 1 42 White G2/3 IVA yes TAX�CARB Complete responseOSPC 2 67 White G3 IIIB yes TAX�CARB Complete responseOSPC 3 61 White G3 IIIC no TAX�CARB Partial responseOSPC 4 60 White G3 IIIC no TAX�CARB Complete responseOSPC 5 59 Afro-American G2/3 IIIC yes TAX�CARB Complete responseOSPC 6 72 White G3 IVA yes TAX�CARB Stable diseaseOSPC 7 63 White G3 IIIC yes TAX�CARB Progressive diseaseOSPC 8 74 Afro-American G2/3 IIIC no TAX�CARB Partial responseOSPC 9 68 White G3 IIIB yes TAX�CARB Complete responseOSPC 10 77 White G2/3 IIIC no TAX�CARB Complete response

15GENE EXPRESSION PROFILES OF OSPC AND NOVA

the samples and were thus most informative. For the hierarchicalclustering shown in Figures 1 and 2, only genes significantlyexpressed and whose average change in expression level was atleast 5-fold were chosen. The expression value of each selectedgene was renormalized to a mean of 0.

Quantitative real-time PCR

Quantitative real-time PCR was performed with an ABI Prism7000 Sequence Analyzer using the manufacturer’s recommendedprotocol (Applied Biosystems, Foster City, CA) to validate differ-ential expression of selected genes in samples from all 15 primarycell lines. Each reaction was run in triplicate. The comparative Ct

method was used to calculate amplification fold as specified by themanufacturer. Briefly, 5 �g of total RNA from each sample werereverse-transcribed using SuperScript II RNAse H Reverse Tran-scriptase (Invitrogen, Carlsbad, CA). Reverse-transcribed RNAsamples (10 �l, from 500 �l of total volume) were amplified usingthe TaqMan Universal PCR Master Mix (Applied Biosystems) toproduce PCR products specific for TROP-1, CD24, claudin 3 andclaudin 4. Primers specific for 18S ribosomal RNA and empiri-cally determined ratios of 18S competimers (Applied Biosystems)were used to control for the amounts of cDNA generated fromeach sample. Primers for TROP-1, claudin 3 and claudin 4 wereobtained from Applied Biosystems as assay on demand products.Assay IDs were Hs00158980_m1 (TROP-1), Hs00265816_s1(claudin 3) and Hs00533616_s1 (claudin 4). CD24 primer se-quences were as follows: forward, 5�-cccaggtgttactgtaattcctcaa;reverse, 5�-gaacagcaatagctcaacaatgtaaac. Differences amongOSPC and NOVA samples in the quantitative real-time PCRexpression data were tested using the Kruskal-Wallis nonparamet-ric test. Pearson product-moment correlations were used to esti-mate the degree of association between the microarray and quan-titative real-time PCR data.

Flow cytometry

To validate microarray data on primary OSPC and NOVA celllines at the protein level, TROP-1/EP-CAM and CD24 expressionwere evaluated by flow-cytometric analysis of a total of 13 primarycell lines (i.e., 10 OSPC and 3 NOVA). Unconjugated anti-TROP-1/EP-CAM (IgG2a), anti-CD24 (IgG2a) and isotype control (mouseIgG2a) antibodies were obtained from BD Pharmingen (San Diego,CA). Goat antimurine FITC-labeled mouse Ig was purchased fromBecton Dickinson (San Jose, CA). Analysis was conducted with aFACScan, utilizing Cell Quest software (Becton Dickinson).

TROP-1 and CD24 immunostaining of formalin-fixed tumortissues

To evaluate whether the differential TROP-1 and CD24 expres-sion detected by flow cytometry in primary OSPC cell lines wascomparable to the expression of TROP-1 and CD24 in unculturedOSPC from which the primary cell lines were derived, proteinexpression was evaluated by standard immunohistochemical stain-ing on formalin-fixed tumor tissue from all surgical specimens(i.e., 10 OSPC and 5 NOVA controls). Study blocks were selectedafter histopathologic review by a surgical pathologist. The mostrepresentative hematoxylin and eosin–stained block sections wereused for each specimen. Briefly, immunohistochemical stains wereperformed on sections (4 �m thick) of formalin-fixed, paraffin-embedded tissue. After pretreatment with 10 mM citrate buffer(pH 6.0) using a steamer, sections were incubated with anti-Ep-CAM Ab-3 and anti-CD24 MAbs (Neo Markers, Fremont, CA) at1:2,000 dilution. Slides were subsequently labeled with strepta-vidin–biotin (Dako, Glostrup, Denmark), stained with diamino-benzidine and counterstained with hematoxylin. The intensity ofstaining was graded as 0 (not greater than negative control), 1�(light), 2� (moderate) or 3� (heavy).

RESULTS

Gene expression profiles distinguish OSPC from NOVA andidentify differentially expressed genes

Flash-frozen biopsies from ovarian tumor tissue contain signif-icant numbers of contaminant stromal cells as well as a variety ofhost-derived immune cells (e.g., monocytes, dendritic cells, lym-phocytes). In addition, because ovarian epithelial cells represent asmall proportion of the total cells found in the normal ovary, it isdifficult to collect primary material that is free of contaminatingovarian stromal cells in sufficient quantities to conduct compara-tive gene expression analyses. However, ovarian epithelial cellscan be isolated and expanded in culture for about 15 passages,3,6

while the majority of primary ovarian carcinomas can be expandedin vitro for several passages.15 Thus, short-term primary OSPC andNOVA cell cultures, minimizing the risk of selection bias inherentin any long-term in vitro growth, may provide an opportunity tostudy differential gene expression between relatively pure popu-lations of normal and tumor-derived epithelial cells. Accordingly,

FIGURE 1 – Hierarchical clustering of 15 primary ovarian cell lines(10 OSPC and 5 NOVA) and 2 established OSPC cell lines (UCI-101and UCI-107).

FIGURE 2 – Molecular profile of 10 primary OSPC and 5 NOVA celllines. Hierarchical clustering of 299 genes with differential expressionbetween 10 OSPC and 5 NOVA groups (p � 0.05) using a 5-foldthreshold. The cluster is color-coded, using red for upregulation, greenfor downregulation and black for median expression. Agglomerativeclustering of genes was illustrated with dendrograms.

16 SANTIN ET AL.

comprehensive gene expression profiles of 10 primary OSPC and5 primary NOVA cell lines were generated using high-densityoligonucleotide arrays with 12,533 probe sets, which in totalincluded some 10,000 genes. In addition, gene expression profilesderived from 2 established and previously characterized OSPC celllines (UCI-101 and UCI-107) were analyzed. By combining thedetection levels of genes significantly expressed in primary andestablished OSPC cultures, we found very little correlation be-tween the 2 groups of OSPC. Indeed, as shown in Figure 1,UCI-101 and UCI-107 cells grouped together in the dendrogram,while all 10 primary OSPCs clustered tightly together in therightmost columns separate from the 5 NOVA controls. Because ofthese results, we focused our analysis on the detection of differ-entially expressed genes between the 2 homogeneous groups ofprimary OSPC and NOVA cell lines. Figure 2 shows the clusteranalysis on hybridization intensity values for each gene in all 15primary cultures of NOVA and OSPC. Using the nonparametricWRS test (p � 0.05), which readily distinguished between the 2groups of primary cultures, we found 1,546 genes differentiallyexpressed between OSPC and NOVA. There were 365 genesshowing a �5-fold change along with “present” detection calls inmore than half the samples in the overexpressed group. Of these,350 were significant by SAM, with a median FDR of 0.35% and a90th percentile FDR of 0.59%. Of the 365 genes, 299 yielded p �0.05 via WRS and 298 were among those found significant bySAM. Figure 2 describes the cluster analysis performed on hybrid-ization intensity values for 298 gene segments whose averagechange in expression level was at least 5-fold and which weresignificant with both the WRS test and SAM. All 10 OSPCs weregrouped together in the rightmost columns. Similarly, in the left-most columns, all 5 NOVA lines clustered tightly together. Thetight clustering of OSPC from NOVA lines was “driven” by 2distinct profiles of gene expression. The first was represented by agroup of 129 genes that were highly expressed in OSPC andunderexpressed in NOVA (Table II). Many genes shown previ-ously to be involved in ovarian carcinogenesis are present on theselists, providing some validity to our array analysis, while others arenovel in ovarian carcinogenesis. Included in these group of genesare laminin, claudin 3, claudin 4, tumor-associated calcium signaltransducers 1 and 2 (TROP-1/Ep-CAM, TROP-2), ladinin 1,S100A2, SERPIN2 (PAI-2), CD24, lipocalin 2, osteopontin, kal-likrein 6 (protease M), kallikrein 10, matriptase (TADG-15) andstratifin (Table II). TROP-1/Ep-CAM, encoding for a transmem-brane glycoprotein previously found to be overexpressed in vari-ous carcinoma types including colorectal and breast21 and whereantibody-directed therapy has resulted in improved survival,22 was39-fold differentially expressed in OSPC compared to NOVA(Table II). The second profile was represented by 170 genes thatwere highly expressed in NOVA and underexpressed in OSPC(Table III). Included in this group of genes are transforminggrowth factor-� receptor III, platelet-derived growth factor recep-tor �, SEMACAP3, ras homolog gene family member I (ARHI),thrombospondin 2 and disabled-2/differentially expressed in ovar-ian carcinoma 2 (Dab2/DOC2) (Table III).

Validation of microarray dataWe used quantitative real-time PCR assays to validate the

microarray data. Four highly differentially expressed genes be-tween OSPC and NOVA (i.e., TROP-1, CD24, claudin-3 andclaudin-4) were selected for quantitative real-time PCR analysis. Acomparison of the microarray and quantitative real-time PCR datafor these genes is shown in Figure 3. Expression differencesbetween OSPC and NOVA for TROP-1 (p � 0.02), CD24 (p �0.004), claudin 3 (p � 0.02) and claudin 4 (p � 0.01) were readilyapparent (Table II, Fig. 3). Moreover, for all 4 genes tested, thequantitative real-time PCR data were highly correlated to themicroarray data (p � 0.001; r � 0.81, 0.90, 0.80 and 0.85, respec-tively), as estimated from the samples (i.e., 10 OSPC and 5 NOVA)included in both the quantitative real-time PCR and microarray ex-periments. Thus, quantitative real-time PCR data suggest that most

array probe sets are likely to accurately measure the levels of theintended transcript within a complex mixture of transcripts.

TROP-1 and CD24 expression by flow cytometry in primaryOSPC and NOVA cell lines

An important issue is whether differences in gene expression resultin meaningful differences in protein expression. Because the TROP-1/Ep-CAM gene encodes the target for the anti-Ep-CAM antibody(17-1A) edrecolomab (Panorex), which has previously been shown toincrease survival in patients harboring stage III colon cancer,22 ex-pression of Ep-CAM protein by FACS analysis was analyzed in 13primary cell lines (10 OSPC and 3 NOVA). As positive controls,breast cancer cell lines (i.e., BT-474 and SK-BR-3; ATCC, Rockville,MD) known to overexpress TROP-1/Ep-CAM were also studied.High TROP-1/Ep-CAM expression was found in all 10 primary OSPCcell lines tested (100% positive cells for all 10 OSPC), with MFIranging 116–280 (Fig. 4). In contrast, primary NOVA cell lines werenegative for TROP-1/Ep-CAM surface expression (p � 0.001) (Fig.4). Similarly, CD24 expression was found in all primary OSPC celllines tested (100% positive cells for all 10 OSPC), with MFI ranging26–55 (Fig. 4). In contrast, primary NOVA cell lines were negativefor CD24 surface expression (p � 0.005) (Fig. 4). These results showthat high expression of the TROP-1/Ep-CAM and CD24 gene prod-ucts by OSPC correlates tightly with high protein expression by tumorcells. Breast cancer-positive controls expressed high levels of TROP-1/Ep-CAM (data not shown).

TROP-1/Ep-CAM and CD24 expression by immunohistology onOSPC and NOVA tissue blocks

To determine whether the high (OSPC) or low (NOVA) expres-sion of the genes and Ep-CAM and CD24 proteins detected bymicroarray and flow cytometry, respectively, in primary cell lineswas the result of selection of a subpopulation of cancer cellspresent in the original tumor or whether in vitro expansion condi-tions may have modified gene expression, we performed immu-nohistochemical analysis of TROP-1/Ep-CAM and CD24 proteinexpression on formalin-fixed tumor tissue from all unculturedprimary surgical specimens of OSPC and NOVA. As shown in theleft panel of Figure 5, heavy apical membranous staining for CD24protein expression was noted in all OSPC specimens that alsooverexpressed the CD24 gene and its gene product by microarrayand flow cytometry, respectively. In contrast, negative or low (i.e.,score 0 or 1�) staining was found in all NOVA samples tested byimmunohistochemistry. Similarly, as shown in the right panel ofFigure 5, heavy membranous staining for TROP-1/Ep-CAM pro-tein expression (i.e., score 3�) was noted in all OSPC specimensthat also overexpressed the TROP-1/Ep-CAM gene and its geneproduct by microarray and flow cytometry, respectively. In con-trast, negative or low (i.e., score 0 or 1�) staining was found in allNOVA samples tested by immunohistochemistry.

DISCUSSION

Because of the lack of an effective ovarian cancer screeningprogram and the common development of chemotherapy-resistantdisease after an initial response to cytotoxic agents (i.e., platinum-based regimen), ovarian cancer remains the most lethal among thegynecologic malignancies.2 Thus, identification of novel ovariantumor markers to be used for early detection of the disease as wellas the development of effective therapy against chemotherapy-resistant/recurrent ovarian cancer remains a high priority.

High-throughput technologies for assaying gene expression,such as high-density oligonucleotide and cDNA microarrays, mayoffer the potential to identify clinically relevant genes highlydifferentially expressed between ovarian tumors and NOVA con-trol epithelial cells.3–11 Our investigation involved genomewideexamination of differences in gene expression between primaryOSPC and NOVA epithelial cells. We used short-term primaryOSPC and NOVA cultures (to minimize the risk of selection biasinherent in any long-term in vitro growth) to study differential

17GENE EXPRESSION PROFILES OF OSPC AND NOVA

gene expression in highly enriched populations of epithelial tumorcells. Only cancer cells derived from papillary serous histology tu-mors, which is the most common histologic type of ovarian cancer,were included, to limit the complexity of gene expression analysis.

We found that hierarchical clustering of samples and geneexpression levels within samples led to the unambiguous separa-tion of OSPC from NOVA. Of interest, the expression patterns

detected in primary OSPC cells were consistently different fromthose seen in established OSPC cell lines (i.e., UCI-101 andUCI-107). These data, thus, further highlight the divergence ofgene expression that occurs as a result of long-term in vitrogrowth. Furthermore, these data emphasize that, although estab-lished ovarian cancer cell lines provide a relatively simple modelto examine gene expression, primary OSPC and NOVA cultures

TABLE II – UPREGULATED GENES EXPRESSED AT LEAST 5 FOLD HIGHER IN OSPC COMPARED WITH NOVA

Probe set Gene symbol Score(d)(SAM) p of WRS Ratio ova/nova Probe set Gene symbol Score(d)(SAM) p of WRS Ratio ova/nova

35280_at LAMC2 1.68927386 0.006 46.4535276_at CLDN4 1.734410451 0.015 43.7633904_at CLDN3 1.650076713 0.02 40.24575_s_at TACSTD1 1.705816336 0.02 39.3632154_at TFAP2A 1.667038647 0.002 33.3139015_f_at KRT6E 1.062629117 0.047 28.021713_s_at CDKN2A 1.137682905 0.015 26.9641376_i_at UGT2B7 0.939735032 0.047 24.8138551_at L1CAM 1.151935363 0.008 24.66291_s_at TACSTD2 1.249487388 0.047 24.4633282_at LAD1 1.422481563 0.006 24.3134213_at KIBRA 1.533570321 0.002 23.0638489_at HBP17 1.522882814 0.004 22.5436869_at PAX8 1.43906836 0.004 22.2038482_at CLDN7 1.307716566 0.027 20.0137909_at LAMA3 1.121654521 0.027 19.2434674_at S100A1 1.219106334 0.008 19.011620_at CDH6 0.908193479 0.036 18.6932821_at LCN2 1.99990601 0.008 18.13522_s_at FOLR3 1.113781518 0.02 17.9039660_at DEFB1 0.837612681 0.036 17.342011_s_at BIK 1.594057668 0.006 17.2341587_g_at FGF18 0.965726983 0.02 17.1036929_at LAMB3 1.115590892 0.047 16.7635726_at S100A2 1.036576352 0.004 15.051887_g_at WNT7A 1.186990893 0.004 14.7535879_at GAL 1.223278825 0.002 14.65266_s_at CD24 1.756569076 0.004 14.451108_s_at EPHA1 1.242309171 0.006 14.3637483_at HDAC9 1.406744957 0.006 14.2831887_at — 1.311220827 0.011 13.681788_s_at DUSP4 1.22421987 0.003 13.6532787_at ERBB3 0.996784565 0.02 13.2141660_at CELSR1 1.634286803 0.004 13.1133483_at NMU 1.100849065 0.004 13.0431792_at ANXA3 0.896090153 0.011 12.9036838_at KLK10 1.026306829 0.02 12.711585_at ERBB3 1.102058608 0.011 12.511898_at TRIM29 1.071987353 0.002 12.4437185_at SERPINB2 0.815945986 0.027 12.26406_at ITGB4 1.296194559 0.006 11.661914_at CCNA1 0.9363427778 0.011 11.21977_s_at CDH1 0.93637461 0.036 11.1937603_at IL1RN 1.103624942 0.015 11.1435977_at DKK1 1.123240701 0.006 10.7436133_at DSP 1.280269127 0.002 10.6936113_s_at TNNT1 1.269558595 0.002 10.191802_s_at ERBB2 0.787465706 0.006 9.612092_s_at SPP1 1.34315986 0.02 9.5335699_at BUB1B 1.026388835 0.006 9.4937554_at KLK6 0.895036336 0.027 9.4538515_at BMP7 0.945367 0.027 9.3234775_at TSPAN-1 1.001195829 0.02 9.0137558_at IMP-3 1.023799379 0.011 8.9938324_at LISCH7 1.308000521 0.006 8.9639610_at HOXB2 1.355268631 0.006 8.64572_at TTK 1.122796615 0.006 8.531970_s_at FGFR2 1.022708001 0.02 8.30160025_at TGFA 1.065272755 0.015 8.2841812_s_at NUP210 1.39287031 0.006 8.2634282_at NFE2L3 1.165273649 0.008 8.062017_s_at CCND1 1.114984456 0.002 8.0433323_r_at SFN 1.202433185 0.008 8.0138766_at SRCAP 1.131917941 0.008 7.9941060_at CCNE1 1.151246634 0.006 7.97

39016_r_at KRT6E 0.973486831 0.008 7.9131610_at MAP17 1.0156502 0.027 7.812027_at S100A2 0.941919001 0.008 7.76418_at MKI67 0.826426448 0.011 7.461536_at CDC6 1.08868941 0.017 7.37634_at PRSS8 0.899891713 0.02 7.3034342_s_at SPP1 1.318723271 0.02 7.27182_at ITPR3 1.107167336 0.006 7.2732382_at UPK1B 0.731294678 0.047 7.16863_g_at SERPINB5 0.783530451 0.015 7.14904_s_at TOP2A 0.971648429 0.02 7.1240095_at CA2 0.798857154 0.027 7.0241294_at KRT7 1.082553892 0.011 7.0039951_at PLS1 0.995091449 0.006 6.9438051_at MAL 0.819842532 0.036 6.8240726_at KIF11 0.803689697 0.036 6.781148_s_at — 0.683569558 0.047 6.7237920_at PITX1 0.996497645 0.015 6.6737117_at ARHGAP8 1.129131077 0.002 6.6538881_i_at TRIM16 0.721698355 0.047 6.5934251_at HOXB5 1.219463307 0.002 6.5241359_at PKP3 1.047269618 0.004 6.5040145_at TOP2A 0.961173129 0.02 6.4837534_at CXADR 0.888147605 0.006 6.3240303_at TFAP2C 0.948734146 0.004 6.3031805_at FGFR3 0.969764101 0.011 6.2833245_at MAPK13 0.877514586 0.011 6.27885_g_at ITGA3 0.702747685 0.036 6.1934693_at STHM 0.872525584 0.008 6.1538555_at DUSP10 0.880305317 0.008 6.1238418_at CCND1 1.071102249 0.002 5.9733730_at RAI3 0.813298748 0.011 5.9039109_at TPX2 1.040973216 0.011 5.8736658_at DHCR24 1.122129795 0.004 5.8135281_at LAMC2 0.747766326 0.047 5.7838749_at MGC29643 0.683275086 0.036 5.771083_s_at MUC1 0.746980491 0.027 5.7540079_at RAI3 0.709840659 0.02 5.732047_s_at JUP 0.815282235 0.011 5.6232275_at SLPI 0.940625784 0.02 5.612020_at CCND1 0.926408163 0.002 5.5133324_s_at CDC2 1.026683994 0.008 5.4736863_at HMMR 0.96343264 0.006 5.461657_at PTPRR 0.764510362 0.02 5.4137985_at LMNB1 0.895475347 0.008 5.3636497_at C14orf78 0.942921564 0.008 5.332021_s_at CCNE1 0.893228297 0.006 5.3337890_at CD47 0.775908217 0.015 5.3340799_at C16orf34 0.852774782 0.008 5.3035309_at ST14 0.852534105 0.008 5.301599_at CDKN3 0.925527261 0.02 5.29981_at MCM4 1.058558782 0.006 5.2832715_at VAMP8 0.938171642 0.006 5.2838631_at TNFAIP2 0.72369235 0.015 5.2634715_at FOXM1 1.31035831 0.008 5.2433448_at SPINT1 0.924028022 0.015 5.21419_at MKI67 0.938133197 0.015 5.161651_at UBE2C 1.436239741 0.008 5.1435769_at GPR56 0.937347548 0.015 5.0837310_at PLAU 0.885110741 0.036 5.0836761_at ZFN339 0.937123503 0.011 5.0537343_at ITPR3 1.001079303 0.003 5.0540425_at EFNA1 0.813414458 0.047 5.041803_at CDC2 0.732852195 0.027 5.00

18 SANTIN ET AL.

TABLE III – UPREGULATED GENES EXPRESSED AT LEAST 5 FOLD HIGHER IN NOVA COMPARED WITH OSPC

Probe set Gene symbol Score(d)(SAM) p of WRS Ratio ova/nova Probe set Gene symbol Score(d)(SAM) p of WRS Ratio ova/nova

39701_at PEG3 1.991111245 0.006 113.3232582_at MYH11 1.921434447 0.002 67.3139673_i_at ECM2 1.740409609 0.011 53.5437394_at C7 1.597329897 0.02 50.4537247_at TCF21 2.261979734 0.002 39.291897_at TGFBR3 1.648143277 0.003 38.1236627_at SPARCL1 1.610346382 0.008 37.8437015_at ALDH1A1 1.886579474 0.002 35.1838469_at TM4SF3 1.620821878 0.003 34.4335717_at ABCA8 1.709820793 0.008 33.9232664_at RNASE4 1.720857082 0.003 32.9440775_at ITM2A 1.393751125 0.006 31.3538519_at NR1H4 1.431579641 0.004 27.0237017_at PLA2G2A 1.263990266 0.011 26.6836681_at APOD 1.44030134 0.008 26.0434193_at CHL1 1.738491852 0.006 25.9734363_at SEPP1 1.490374268 0.015 25.931501_at IGF1 1.116943817 0.027 25.8733240_at SEMACAP3 1.818843975 0.003 25.5436939_at GPM6A 0.924236354 0.047 25.47614_at PLA2G2A 1.391395227 0.003 23.1537407_s_at MYH11 1.72766007 0.002 22.7339325_at EBAF 1.248164036 0.02 22.49767_at — 1.688001805 0.002 21.9037595_at — 1.582101386 0.004 20.941290_g_at GSTM5 1.383630361 0.003 20.8434388_at COL14A1 1.400078214 0.015 20.39607_s_at VWF 1.314435559 0.002 19.0537599_at AOX1 1.669903577 0.003 17.6141504_s_at MAF 1.463988429 0.008 16.4041412_at PIPPIN 1.799353403 0.002 16.08279_at NR4A1 1.194733065 0.008 15.4238427_at COL15A1 1.570514035 0.002 15.3841405_at SFRP4 1.478603828 0.002 14.4439066_at MFAP4 1.91469237 0.004 14.261731_at PDGFRA 1.791307012 0.003 13.9136595_s_at GATM 1.382271028 0.004 13.8634343_at STAR 2.080476608 0.003 13.6736917_at LAMA2 1.359731285 0.006 13.5138430_at FABP4 1.054221974 0.02 13.0536596_r_at GATM 1.22177547 0.008 12.6735898_at WISP2 1.276226302 0.004 12.5536606_at CPE 1.608244463 0.003 12.3032057_at LRRC17 1.345223643 0.011 12.2233431_at FMOD 1.516795166 0.003 12.1734985_at CILP 0.905018335 0.02 11.53755_at ITPR1 1.433938835 0.002 11.061466_s_at FGF7 1.184028604 0.027 11.0036727_at — 0.98132702 0.036 10.961103_at RNASE4 1.456068199 0.002 10.8832666_at CXCL12 1.342426238 0.006 10.72914_g_at ERG 1.264721284 0.002 10.5440698_at CLECSF2 1.325237675 0.002 10.4636873_at VLDLR 1.344197327 0.004 10.451090_f_at — 0.914708216 0.027 10.3436042_at NTRK2 0.950553444 0.02 10.3236311_at PDE1A 1.356950738 0.004 10.2141685_at NY-REN-7 0.8848466 0.036 10.0832847_at MYLK 1.545610138 0.002 10.0035358_at TENC1 1.539140855 0.003 9.9732249_at HFL1 1.257702238 0.02 9.8636695_at na 1.452847153 0.003 9.821987_at PDGFRA 1.50655467 0.002 9.7637446_at GASP 1.219014593 0.004 9.7635752_s_at PROS1 1.211272096 0.008 9.6636533_at PTGIS 1.882348646 0.004 9.6238886_i_at ARHI 1.127672988 0.02 9.5936733_at FLJ32389 1.420588897 0.011 9.5738717_at DKFZP586A0522 1.158933663 0.015 9.5032551_at EFEMP1 1.385495033 0.004 9.381968_g_at PDGFRA 1.364848071 0.003 9.3133910_at PTPRD 1.129963902 0.008 9.2032778_at ITPR1 1.370809534 0.002 9.08280_g_at NR4A1 1.074894321 0.006 8.7935389_s_at ABCA6 1.209294071 0.011 8.7932889_at RPIB9 1.145333813 0.003 8.7437248_at CPZ 1.238797022 0.002 8.6939674_r_at ECM2 0.874009817 0.027 8.6733911_at PTPRD 1.099609918 0.02 8.6635234_at RECK 1.407865518 0.008 8.5832119_at — 1.153957574 0.011 8.5735998_at LOC284244 1.104281231 0.008 8.5437279_at GEM 1.012760866 0.008 8.3135702_at HSD11B1 1.164189513 0.004 8.2832126_at FGF7 1.336918337 0.008 8.22

36867_at — 1.273166453 0.008 8.2138653_at PMP22 1.422063697 0.002 8.1938875_r_at GREB1 1.026886865 0.015 8.1035366_at NID 1.483421362 0.002 8.1034417_at FLJ36166 0.783978445 0.047 7.9837221_at PRKAR2B 0.927090765 0.036 7.9139031_at COX7A1 1.564725491 0.004 7.8939757_at SDC2 1.288106392 0.002 7.8036629_at DSIPI 0.981563882 0.008 7.7935390_at ABCA6 1.026714913 0.036 7.7939629_at PLA2G5 1.405181995 0.002 7.7040961_at SMARCA2 0.996692724 0.015 7.68719_g_at PRSS11 1.399043078 0.002 7.6540856_at SERPINF1 1.077533093 0.008 7.5537008_r_at SERPINA3 1.134224016 0.002 7.5333834_at CXCL12 1.060878451 0.002 7.5131880_at D8S2298E 1.177864913 0.002 7.4537628_at MAOB 1.194963489 0.004 7.4334853_at FLRT2 1.250330254 0.027 7.4138887_r_at ARHI 1.169953614 0.015 7.3238220_at DPYD 1.024334451 0.02 7.261327_s_at MAP3K5 0.891703475 0.02 7.231380_at FGF7 1.096254206 0.004 7.1437573_at ANGPTL2 1.052539345 0.002 7.08718_at PRSS11 1.381205346 0.002 6.9936712_at — 1.15195149 0.005 6.881709_g_at MAPK10 1.160327795 0.002 6.8539123_s_at TRPC1 1.060327922 0.015 6.7938627_at HLF 0.911787462 0.036 6.7932076_at DSCR1L1 1.127515982 0.002 6.7736669_at FOSB 1.023057503 0.011 6.6538194_s_at IGKC 1.239936045 0.015 6.6439545_at CDKN1C 1.040717569 0.004 6.6236993_at PDGFRB 1.384657766 0.004 6.6035837_at SCRG1 1.023840456 0.036 6.481507_s_at EDNRA 1.23933124 0.004 6.4840488_at DMD 1.291791538 0.002 6.4238364_at — 1.030881108 0.004 6.3541424_at PON3 0.946224951 0.036 6.3232109_at FXYD1 1.005577422 0.004 6.191182_at PLCL1 1.097390316 0.002 6.1731897_at DOC1 1.533672652 0.003 6.1337208_at PSPHL 1.007759699 0.015 6.0836396_at — 1.009684807 0.015 6.0741505_r_at MAF 1.116101319 0.006 6.0637765_at LMOD1 1.127716375 0.003 6.0037398_at PECAM1 0.970664041 0.008 5.9841013_at FLJ31737 1.036561659 0.003 5.9839279_at BMP6 1.106724571 0.002 5.931527_s_at CG018 0.804755548 0.047 5.9139038_at FBLN5 1.279283798 0.004 5.8932542_at FHL1 1.134214637 0.002 5.8838508_s_at TNXB 0.878513741 0.011 5.7432696_at PBX3 0.888011703 0.027 5.6941796_at PLCL2 0.857601993 0.02 5.6834473_at TLR5 0.871815246 0.027 5.67661_at GAS1 1.267909476 0.004 5.6641449_at SGCE 1.050056933 0.004 5.6535740_at EMILIN1 1.366368794 0.011 5.5837908_at GNG11 0.989043327 0.004 5.4337406_at MAPRE2 1.265872665 0.002 5.4133802_at HMOX1 1.034088234 0.015 5.4139106_at APOA1 1.266005754 0.008 5.401771_s_at PDGFRB 1.336082701 0.006 5.3939409_at C1R 1.05784087 0.011 5.3932535_at FBN1 1.422415283 0.006 5.3537710_at MEF2C 0.98149558 0.011 5.3537958_at TM4SF10 1.293658009 0.003 5.3533756_at AOC3 0.829203515 0.02 5.2936569_at TNA 0.926096917 0.006 5.2539771_at RHOBTB1 1.048906896 0.008 5.2039852_at SPG20 0.82401517 0.027 5.2035168_f_at COL16A1 1.509830282 0.011 5.1833244_at CHN2 0.92878389 0.015 5.1835681_r_at ZFHX1B 1.170745794 0.006 5.142087_s_at CDH11 1.656534188 0.008 5.1240496_at C1S 0.973175912 0.011 5.1041137_at PPP1R12B 1.12885067 0.008 5.0739698_at HOP 0.797252583 0.011 5.0538211_at ZNF288 0.926263264 0.015 5.0441839_at GAS1 1.127093791 0.006 5.0339979_at F10 0.890787173 0.002 5.021135_at GPRK5 1.150554994 0.002 5.01479_at DAB2 1.255638531 0.006 5.01

19GENE EXPRESSION PROFILES OF OSPC AND NOVA

represent a better model system of normal and cancerous ovariantissues in comparative gene expression analysis. Because of theseresults, we focused our analysis on the detection of differentiallyexpressed genes between the 2 homogeneous groups of primaryOSPC and NOVA.

We detected 298 genes differentially expressed between OSPCand NOVA whose average change in expression level between the2 groups was at least 5-fold and which were significant with both

the WRS test and SAM. The known function of some of thesegenes may provide insights in the biology of serous ovariantumors, while others may prove to be useful diagnostic and ther-apeutic markers against OSPC. For example, the laminin �2 genewas the most highly differentially expressed gene in OSPC, withover 46-fold upregulation relative to NOVA. Migration of ovarianepithelial cells is considered essential for cell dissemination andinvasion of the submesothelial extracellular matrix commonly seen

FIGURE 3 – Quantitative real-time PCR and microarray expres-sion analyses of TROP-1, CD24,claudin 3 and claudin 4 genes dif-ferentially expressed betweenOSPC and NOVA.

20 SANTIN ET AL.

in ovarian cancer. Consistent with this view, the laminin �2isoform has been previously suggested to play an important role inthe adhesion, migration and scattering of ovarian carcinomacells.23–25 Thus, it is likely that the high laminin expression foundin ovarian tumor cells may be a marker correlated with the inva-sive potential of OSPC. Consistent with this view, increased cellsurface expression of laminin has been reported in highly meta-static tumors cells compared to cells of low metastatic potential.26

Previous work has also shown that attachment and metastasis oftumor cells can be inhibited by incubation with antilaminin anti-bodies27 or synthetic laminin peptides,28 underscoring the novelpotential for the treatment of chemotherapy-resistant ovarian can-cer.

TROP-1/Ep-CAM (also called 17-1A, ESA, EGP40) is a 40 kDaepithelial transmembrane glycoprotein overexpressed in severalnormal epithelia and in various carcinomas, including colorectal

and breast cancers.21 In most adult epithelial tissues, enhancedexpression of Ep-CAM is closely associated with either benign ormalignant proliferation. Among mammals, Ep-CAM is an evolu-tionarily highly conserved molecule,29 suggesting an importantbiologic function in epithelial cells and tissue. In this regard,Ep-CAM is known to function as an intercellular adhesion mole-cule and could have a role in tumor metastasis.30 Because arandomized phase II trial with MAb CO17-1A in colorectal car-cinoma patients has demonstrated a significant decrease in recur-rence and mortality of MAb-treated patients vs. controls,22 TROP-1/Ep-CAM antigen has attracted substantial attention as a targetfor immunotherapy in human carcinomas. In our work, TROP-1/Ep-CAM was 39-fold differentially expressed in OSPC comparedto NOVA. These data support the notion that anti-Ep-CAM anti-body therapy may be a novel and potentially effective treatmentoption for OSPC patients with residual/resistant disease after sur-

FIGURE 4 – Representative FACSanalysis of CD24 staining (leftpanel) and TROP-1/Ep-CAM stain-ing (right panel) of 2 primaryOSPC cell lines and 1 NOVA cellline. Data on CD24 and TROP-1/Ep-CAM are shown in solid black,while isotype control MAb profilesare shown in white. Expression ofboth CD24 and TROP-1/Ep-CAMwas significantly higher in OSPCcompared to NOVA cell lines (p �0.001 by Student’s t-test).

21GENE EXPRESSION PROFILES OF OSPC AND NOVA

gical and cytotoxic therapy. Protein expression data obtained byflow cytometry on primary OPSC cell lines and by immunohisto-chemistry on uncultured OSPC blocks support this view.

Claudin 3 and claudin 4, 2 members of claudin family of tightjunction proteins, were 2 of the top 5 differentially expressed genesin OSPC. These results are consistent with a previous report on

gene expression in ovarian cancer.6 Although the function ofclaudin proteins in ovarian cancer remains unclear, they likelyrepresent a transmembrane receptor.31 Of interest, claudin 3 andclaudin 4 are homologous to CPE-R, the low- and high-affinityintestinal epithelial receptors for CPE, respectively, and are suffi-cient to mediate CPE binding and trigger subsequent toxin-medi-

FIGURE 5 – Representative immunohistochemical staining for CD24 (left panel) and Trop-1/Ep-CAM (right panel) markers of 2 paraffin-embedded OSPC specimens and 1 NOVA specimen. NOVA 1 (upper panel, right and left) showed negative or light (1�) staining for both CD24and Trop-1/Ep-CAM, while OSPC 1 and OSPC 3 showed heavy apical membranous staining for CD24 (lower and middle panel, left) and strongmembranous staining for TROP-1/Ep-CAM (right). Original magnification �400.

22 SANTIN ET AL.

ated cytolysis.32 These known functions of claudin 3 and claudin4 combined with their extremely high level of expression in OSPChave suggested the use of CPE as a novel therapeutic strategy totreat chemotherapy-resistant disease in ovarian cancer patients.Supporting this view, the functional cytotoxicity of CPE in meta-static, androgen-independent prostate cancer overexpressing clau-din 3 has been reported.33

PAI-2, a gene whose expression has been linked to cell invasionin several human malignancies34,35 as well as to protection fromTNF-�-mediated apoptosis,36 was 12-fold differentially expressedin OSPC compared to NOVA. Consistent with our findings, pre-vious studies have shown that elevated levels of PAI-2 are detect-able in the ascites of ovarian cancer patients and that high PAI-2levels are independently predictive of poor disease-free surviv-al.37,38 In some of these studies, a 7-fold increase in PAI-2 contentwas found in the omentum of ovarian cancer patients compared toprimary disease, suggesting that metastatic tumors may overex-press PAI-2.38 Other studies, however, have identified PAI-2 pro-duction as a favorable prognostic factor in epithelial ovarian can-cer.39 Indeed, high PAI-2 expression in invasive ovarian tumorswas limited to a group of OSPC patients who experienced moreprolonged disease-free and overall survival.39 The reasons forthese differences are not clear but, as previously suggested,40 maybe related at least in part to the actions of macrophage colonystimulating factor-1, a cytokine which stimulates the release ofPAI-2 by ovarian cancer cells.

CD24 is a small, heavily glycosylated glycosylphosphatidyl-inositol-linked cell surface protein, which is expressed in hemato-logic malignancies as well as in a large variety of solid tu-mors.41–46 However, it is only recently that CD24 overexpressionhas been reported at the RNA level in ovarian cancer.4 Consistentwith this report, we found the CD24 gene to be 14-fold differen-tially expressed in OSPC compared to NOVA. Because CD24 is aligand of P-selectin, an adhesion receptor on activated endothelialcells and platelets, its expression may contribute to the metastaticcapacities of CD24-expressing ovarian tumor cells.47–49 CD24expression has been reported to be an independent prognosticmarker for ovarian cancer patient survival.50 These data combinedwith our findings further suggest that this marker may delineateaggressive ovarian cancer disease and may be targeted for thera-peutic and/or diagnostic purposes.

Among the genes we have identified, to the best of our knowl-edge, lipocalin 2 has not been previously linked to ovarian cancer.Lipocalin 2 represents a particularly interesting marker because ofseveral features. Lipocalins are extracellular carriers of lipophilicmolecules, such as retinoids, steroids and fatty acids, all of whichmay play important roles in the regulation of epithelial cellgrowth.51,52 In addition, because lipocalin is a secreted protein, itmay play a role in the regulation of cell proliferation and sur-vival.51,52 Of interest, 2 publications, on gene expression profilingof breast and pancreatic cancers, have proposed lipocalin 2 as anovel therapeutic and diagnostic marker for prevention and treat-ment of these diseases.53,54 On the basis of our findings, lipocalin2 may be added to the known markers for ovarian cancer.

Osteopontin (SPP1) is an acidic, calcium-binding glycophos-phoprotein that has been linked to tumorigenesis in several exper-imental animal models and human patient studies.55–57 Because of

its integrin-binding, arginine-glycine-aspartate domain and adhe-sive properties, osteopontin has been reported to play a crucial rolein the metastatic process of several human tumors.55,58 However, itis only recently that the upregulated expression of osteopontin inovarian cancer has been identified.59 Because of the secretednature of this protein, osteopontin has been proposed as a novelbiomarker for the early recognition of ovarian cancer.59 In ourstudy, the SPP1 gene was 10-fold differentially expressed in OSPCcompared to NOVA. Taken together, these data confirm a highexpression of osteopontin in OSPC and further suggest that futureresearch assessing its clinical usefulness in ovarian cancer wouldbe worthwhile.

The organization of kallikreins, a gene family consisting of 15genes that encode for trypsin-like or chymotrypsin-like serineproteases, has been elucidated.60 Serine proteases have well-char-acterized roles in diverse cellular activities, including blood coag-ulation, wound healing, digestion and immune responses, as wellas tumor invasion and metastasis (reviewed by Diamandis andYousef60). Because of the secreted nature of some of these en-zymes, prostate-specific antigen and kallikrein 2 have alreadyfound important clinical application as prostate cancer biomark-ers.60 Of interest, kallikrein 6 (also known as zyme/proteaseM/neurosin), kallikrein 10 and matriptase (TADG-15/MT-SP1)were highly differentially expressed in OSPC compared to NOVA.These data confirm previous results from our group as well asothers showing high expression of several kallikrein genes andproteins in ovarian neoplasms.60–64 Moreover, these results, ob-tained by high-throughput technologies for assaying gene expres-sion, further emphasize the view that some members of the kal-likrein family have the potential to become novel cancer markersfor early diagnosis of ovarian cancer64 as well as targets for noveltherapies against recurrent/refractory ovarian disease.65 Otherhighly ranked genes in OSPC included stratifin, desmoplakin,S100A2, cytokeratin 6, cytokeratin 7 and MUC-1.

A large number of downregulated (at least 5-fold) genes inOSPC vs. NOVA, such as transforming growth factor beta recep-tor III, platelet-derived growth factor receptor �, SEMACAP3, rashomolog gene family member I (ARHI), thrombospondin 2 anddisabled-2/differentially expressed in ovarian carcinoma 2 (Dab2/DOC2) (Table III), were identified in our analysis. Some encodefor widely held tumor-suppressor genes, such as SEMACAP3,ARHI and Dab2/DOC2,66 and others for proteins important forovarian tissue homeostasis or previously implicated in apoptosis,proliferation, adhesion or tissue maintenance. Because of spacelimitations, we will not comment further on the cluster of genesthat showed downregulation of the transcripts in invasive tumors.

In conclusion, several OSPC-restricted markers have been iden-tified through our analysis. Some of these genes have been previ-ously reported to be highly expressed in ovarian cancer, whileothers have not been linked with this disease. Finally, identifica-tion of TROP-1/Ep-CAM and CPE epithelial receptors as some ofthe most highly differentially expressed genes in OSPC comparedto NOVA suggests that novel therapeutic strategies targetingTROP-1/Ep-CAM by MAbs22 and/or claudin 3 and claudin 4 bylocal and/or systemic administration of CPE33 may represent noveltherapeutic modalities in patients harboring OSPC refractory tostandard treatment.

REFERENCES

1. Jemal A, Murray T, Samuels A, Ghafoor A, Ward E, Thun MJ.Cancer statistics. CA Cancer J Clin 2003;53:5–26

2. DiSaia PJ, Creasman WT, eds. Epithelial ovarian cancer. Clinicalgynecologic oncology. St. Louis: Mosby Year Book, 1997. 282–350.

3. Ismail RS, Baldwin RL, Fang J, Browning D, Karlan BY, Gasson JC,Chang DD. Differential gene expression between normal and tumor-derived ovarian epithelial cells. Cancer Res 2000;60:6744–9.

4. Welsh JB, Zarrinkar PP, Sapinoso LM, Kern SG, Behling CA, MonkBJ, Lockhart DJ, Burger RA, Hampton GM. Analysis of gene expres-sion profiles in normal and neoplastic ovarian tissue samples identifies

candidate molecular markers of epithelial ovarian cancer. Proc NatlAcad Sci USA 2001;98:1176–81.

5. Schwartz DR, Kardia SL, Shedden KA, Kuick R, Michailidis G,Taylor JM, Misek DE, Wu R, Zhai Y, Darrah DM, Reed H, EllensonLH, et al. Gene expression in ovarian cancer reflects both morphologyand biologic behavior, distinguishing clear cell from other poor-prognosis ovarian carcinomas. Cancer Res 2002; 62:4722–9.

6. Hough CD, Sherman-Baust CA, Pizer ES, Montz FJ, Im DD, Rosen-shein NB, Cho, KR, Riggins GJ, Morin PJ. Large-scale serial analysisof gene expression reveals genes differentially expressed in ovariancancer. Cancer Res 2000;60:6281–7.

23GENE EXPRESSION PROFILES OF OSPC AND NOVA

7. Ono K, Tanaka T, Tsunoda T, Kitahara O, Kihara C, Okamoto A,Ochiai, K, Takagi T, Nakamura Y. Identification by cDNA microarrayof genes involved in ovarian carcinogenesis. Cancer Res 2000;60:5007–11.

8. Shridhar V, Lee J, Pandita A, Iturria S, Avula R, Staub J, MorrisseyM, Calhoun E, Sen A, Kalli K, Keeney G, Roche P, et al. Geneticanalysis of early- versus late-stage ovarian tumors. Cancer Res 2001;61:5895–904.

9. Hough CD, Cho KR, Zonderman AB, Schwartz DR, Morin PJ.Coordinately upregulated genes in ovarian cancer. Cancer Res 2001;6:3869–76.

10. Shridhar V, Sen A, Chien J, Staub J, Avula R, Kovats S, Lee J, LillieJ, Smith DI. Identification of underexpressed genes in early- andlate-stage primary ovarian tumors by suppression subtraction hybrid-ization. Cancer Res 2002;62:262–70.

11. Jazaeri AA, Lu K, Schmandt R, Harris CP, Rao PH, Sotiriou C,Chandramouli GV, Gershenson DM, Liu ET. Molecular determinantsof tumor differentiation in papillary serous ovarian carcinoma. MolCarcinog 2003;36:53–9.

12. Rosenwald A, Wright G, Chan WC, Connors JM, Campo E, FisherRI, Gascoyne RD, Muller-Hermelink HK, Smeland EB, Giltnane JM,Hurt EM, Zhao H, et al. Lymphoma/Leukemia Molecular ProfilingProject. The use of molecular profiling to predict survival after che-motherapy for diffuse large-B-cell lymphoma. N Engl J Med 2002;346:1937–47.

13. Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, HastieT, Eisen MB, Van de Rijn M, Jeffrey SS, Thorsen T, Quist H, et al.Gene expression patterns of breast carcinomas distinguish tumorsubclasses with clinical implications. Proc Natl Acad Sci USA 2001;98:10869–74.

14. Zorn KK, Jazaeri AA, Awtrey CS, Gardner GJ, Mok SC, Boyd J,Birrer MJ. Choice of normal ovarian control influences determinationof differentially expressed genes in ovarian cancer expression profil-ing studies. Clin Cancer Res 2003;9:4811–8.

15. Santin AD, Ravaggi A, Bellone S, Pecorell, S, Parham GP, CannonMJ. Induction of ovarian tumor CD8� cytotoxic T lymphocytes byacid-eluted peptide pulsed autologous dendritic cells. Obstet Gynecol2000;96:422–30.

16. Fuchtner C, Emma DA, Manetta A, Gamboa G, Bernstein R, Liao SY.Characterization of a human ovarian carcinoma cell line: UCI 101.Gynecol Oncol 1993;48:203–9.

17. Gamboa G, Carpenter PM, Podnos YD, Dorion G, Iravani L, BoltonD, Mascarello JT, Manetta A. Characterization and development ofUCI 107, a primary human ovarian carcinoma cell line. GynecolOncol 1995;58:336–43.

18. Zhan F, Tian E, Bumm K, Smith R, Barlogie B, Shaughnessy J Jr.Gene expression profiling of human plasma cell differentiation andclassification of multiple myeloma based on similarities to distinctstages of late-stage B-cell development. Blood 2002;99:1745–57.

19. Tusher VG, Tibshirani R, Chu G. Significance analysis of microarraysapplied to the ionizing radiation response. Proc Natl Acad Sci USA2001;98:5116–21.

20. Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysis anddisplay of genome-wide expression patterns. Proc Natl Acad Sci USA1998;95:14863–8.

21. Packeisen J, Kaup-Franzen C, Knieriem HJ. Detection of surfaceantigen 17-1A in breast and colorectal cancer. Hybridoma 1999;18:37–40.

22. Riethmuller G, Holz E, Schlimok G, Schmiegel W, Raab R, HoffkenK, Gruber R, Funke I, Pichlmaier H, Hirche H, Buggisch P, Witte J,et al. Monoclonal antibody therapy for resected Dukes’ C colorectalcancer: seven-year outcome of a multicenter randomized trial. J ClinOncol 1998;16:1788–94.

23. Maatta M, Virtanen I, Burgeson R, Autio-Harmainen H. Comparativeanalysis of the distribution of laminin chains in the basement mem-branes in some malignant epithelial tumors: the alpha 1 chain oflaminin shows a selected expression pattern in human carcinomas.J Histochem Cytochem 2001;49:711–26.

24. Casey RC, Skubitz AP. CD44 and beta1 integrins mediate ovariancarcinoma cell migration toward extracellular matrix proteins. ClinExp Metastasis 2000;18:67–75.

25. Yoshida Y, Hosokawa K, Dantes A, Kotsuji F, Kleinman HK, Am-sterdam A. Role of laminin in ovarian cancer tumor growth andmetastasis via regulation of Mdm2 and Bcl-2 expression. Int J Oncol2001;18:913–21.

26. Koshikawa N, Moriyama K, Takamura H, Mizushima H, NagashimaY, Yanoma S, Miyazaki K. Overexpression of laminin gamma 2 chainmonomer in invading gastric carcinoma cells. Cancer Res 1999;59:5596–601.

27. Vollmers HP, Imhof BA, Braun S, Waller CA, Schirrmacher V,Birchmeier W. Monoclonal antibodies which prevent experimentallung metastases. Interference with the adhesion of tumour cells tolaminin. FEBS Lett 1984;172:17–20.

28. Iwamoto Y, Robey FA, Graf J, Sasaki M, Kleinman HK, Yamada Y,Martin GR. YIGSR, a synthetic laminin pentapeptide, inhibits exper-imental metastasis formation. Science 1987;238:1132–4.

29. Linnenbach AJ, Seng BA, Wu S, Robbins S, Scollon M, Pyrc JJ,Druck T, Huebner K. Retroposition in a family of carcinoma-associ-ated antigen genes. Mol Cell Biol 1993;13:1507–15.

30. Litvinov SV, Velders MP, Bakker HA, Fleuren GJ, Warnaar SO.Ep-CAM: a human epithelial antigen is a homophilic cell–cell adhe-sion molecule. J Cell Biol 1994;125:437–46.

31. Morita K, Furuse M, Fujimoto K, Tsukita S. Claudin multigene familyencoding four-transmembrane domain protein components of tightjunction strands. Proc Natl Acad Sci USA 1999;96:511–6.

32. McClane BA. The complex interactions between Clostridium perfrin-gens enterotoxin and epithelial tight junctions. Toxicon 2001;39:1781–91.

33. Long H, Crean CD, Lee WH, Cummings OW, Gabig TG. Expressionof Clostridium perfringens enterotoxin receptors claudin 3 and clau-din 4 in prostate cancer epithelium. Cancer Res 2001;61:7878–81.

34. Ganesh S, Sier CF, Griffioen G, Vloedgraven HJ, de Boer A, WelvaartK, van de Velde CJ, van Krieken JH, Verheijen JH, Lamers CB,Lamers CBHW. Prognostic relevance of plasminogen activators andtheir inhibitors in colorectal cancer. Cancer Res 1995;54:4065–71.

35. Hasina R, Hulett K, Bicciato S, Di Bello C, Petruzzelli GJ, LingenMW. Plasminogen activator inhibitor-2: a molecular biomarker forhead and neck cancer progression. Cancer Res 2003;63:555–9.

36. Dickinson JL, Bates EJ, Ferrante A, Antalis TM. Plasminogen acti-vator inhibitor type 2 inhibits tumor necrosis factor alpha-inducedapoptosis. Evidence for an alternate biologic function. J Biol Chem1995;270:27894–904.

37. Kuhn W, Pache L, Schmalfeldt B, Dettmar P, Schmitt M, Janicke F,Graeff H. Urokinase (uPA) and PAI-1 predict survival in advancedovarian cancer patients (FIGO III) after radical surgery and platinum-based chemotherapy. Gynecol Oncol 1994;5:401–9.

38. Schmalfeldt B, Kuhn W, Reuning U, Pache L, Dettmar P, Schmitt M,Janicke F, Hofler H, Graeff H. Primary tumor and metastasis inovarian cancer differ in their content of urokinase-type plasminogenactivator, its receptor, and inhibitors types 1 and 2. Cancer Res1995;55:3958–63.

39. Chambers SK, Ivins CM, Carcangiu ML. Expression of plasminogenactivator inhibitor-2 in epithelial ovarian cancer: a favorable prognos-tic factor related to the actions of CSF-1. Int J Cancer 1997;74:71–5.

40. Chambers SK, Kacinski BM, Ivins CM, Carcangiu ML. Overexpres-sion of epithelial macrophage colony-stimulating factor (CSF-1) andCSF-1 receptor: a poor prognostic factor in epithelial ovarian cancer,contrasted with a protective effect of stromal CSF-1. Clin Cancer Res1997;3:999–1007.

41. Droz D, Zachar D, Charbit L, Gogusev J, Chretien Y, Iris L. Expres-sion of the human nephron differentiation molecules in renal cellcarcinoma. Am J Pathol 1990;137:895–905.

42. Raife TJ, Lager DJ, Kemp JD, Dick FR. Expression of CD24 (BA-1)predicts monocytic lineage in acute myeloid leukemia. Am J ClinPathol 1994;101:296–9.

43. Jackson D, Waibel R, Weber E, Bell J, Stahel RA. CD24, a signal-transducing molecule expressed on human B cells, is a major surfaceantigen on small cell lung carcinomas. Cancer Res 1992;52:5264–70.

44. Huang LR, Hsu HC. Cloning and expression of CD24 gene in humanhepatocellular carcinoma: a potential early tumor marker gene corre-lates with p53 mutation and tumor differentiation. Cancer Res 1995;55:4717–21.

45. Senner V, Sturm A, Baur I, Schrell UH, Distel L, Paulus W. CD24promotes invasion of glioma cells in vivo. J Neuropathol Exp Neurol1999;58:795–802.

46. Fogel M, Friederichs J, Zeller Y, Husar M, Smirnov A, Roitman L,Altevogt P, Sthoeger ZM. CD24 is a marker for human breast carci-noma. Cancer Lett 1999;143:87–94.

47. Aigner S, Sthoeger ZM, Fogel M, Weber E, Zarn J, Ruppert M, ZellerY, Vestweber D, Stahe R, Sammar M, Altevogt P. CD24, a mucin-type glycoprotein, is a ligand for P-selectin on human tumor cells.Blood 1997;89:3385–95.

48. Aigner S, Ramos CL, Hafezi-Moghadam A, Lawrence MB, Fried-erichs J, Altevogt P, Ley K. CD24 mediates rolling of breast carci-noma cells on P-selectin. EMBO J 1998;12:1241–51.

49. Friederichs J, Zeller Y, Hafezi-Moghadam A, Grone HJ, Ley K,Altevogt P. The CD24/P-selectin binding pathway initiates lung arrestof human A125 adenocarcinoma cells. Cancer Res 2000;60:6714–22.

50. Kristiansen G, Denkert C, Schluns K, Dahl E, Pilarsky C, HauptmannS. CD24 is expressed in ovarian cancer and is a new independentprognostic marker of patient survival. Am J Pathol 2002;161:1215–21.

51. Bratt T. Lipocalins and cancer. Biochim Biophys Acta 2000;1482:318–26.

52. Flower DR. The lipocalin protein family: structure and function.Biochem J 1996;318:1–14.

24 SANTIN ET AL.

53. Terris B, Blaveri E, Crnogorac-Jurcevic T, Jones M, Missiaglia E,Ruszniewski P, Sauvanet A, Lemoine NR. Characterization of geneexpression profiles in intraductal papillary-mucinous tumors of thepancreas. Am J Pathol 2002;160:1745–54.

54. Seth P, Porter D, Lahti-Domenici J, Geng Y, Richardson A, Polyak K.Cellular and molecular targets of estrogen in normal human breasttissue. Cancer Res 2002;62:4540–4.

55. Oates AJ, Barraclough R, Rudland PS. The role of osteopontin intumorigenesis and metastasis. Invasion Metastasis 1997;17:1–15.

56. Koeneman KS, Yeung F, Chung LW. Osteomimetic properties ofprostate cancer cells: a hypothesis supporting the predilection ofprostate cancer metastasis and growth in the bone environment. Pros-tate 1999;39:246–61.

57. Chambers AF, Wilson SM, Kerkvliet N, O’Malley FP, Harris JF,Casson AG. Osteopontin expression in lung cancer. Lung Cancer1996;15:311–23.

58. Denhardt DT, Mistretta D, Chambers AF, Krishna S, Porter JF,Raghuram S, Rittling SR. Transcriptional regulation of osteopontinand the metastatic phenotype: evidence for a Ras-activated en-hancer in the human OPN promoter. Clin Exp Metastasis 2003;20:77– 84.

59. Kim JH, Skates SJ, Uede T, Wong KK, Schorge JO, Feltmate CM,Berkowitz RS, Cramer DW, Mok SC. Osteopontin as a potentialdiagnostic biomarker for ovarian cancer. JAMA 2002;287:1671–9.

60. Diamandis EP, Yousef GM. Human tissue kallikreins: a family of newcancer biomarkers. Clin Chem 2002;48:1198–205.

61. Tanimoto H, Underwood LJ, Shigemasa K, Yan Yan MS, Clarke J,Parmley TH, O’Brien TJ. The stratum corneum chymotryptic enzymethat mediates shedding and desquamation of skin cells is highlyoverexpressed in ovarian tumor cells. Cancer 1999;86:2074–82.

62. Tanimoto H, Underwood LJ, Wang Y, Shigemasa K, Parmley TH,O’Brien TJ. Ovarian tumor cells express a transmembrane serineprotease: a potential candidate for early diagnosis and therapeuticintervention. Tumor Biol 2001;22:104–14.

63. Tanimoto H, Underwood LJ, Shigemasa K, Parmley TH, O’BrienTJ. Increased expression of protease M in ovarian tumors. Tumor Biol2001;22:11–8.

64. Diamandis EP, Scorilas A, Fracchioli S, Van Gramberen M, DeBruijn H, Henrik A, Soosaipillai A, Grass L, Yousef GM, StenmanUH, Massobrio M, Van Der Zee AG, et al. Human kallikrein 6 (hK6):a new potential serum biomarker for diagnosis and prognosis ofovarian carcinoma. J Clin Oncol 2003;21:1035–43.

65. Cannon MJ, O’Brien TJ, Underwood LJ, Crew MD, Bondurant KL,Santin AD. Novel target antigens for dendritic cell–based immuno-therapy against ovarian cancer. Expert Rev Anticancer Ther 2002;2:97–105.

66. Liu Y, Ganesan TS. Tumour suppressor genes in sporadic epithelialovarian cancer. Reproduction 2002;123:341–53.

25GENE EXPRESSION PROFILES OF OSPC AND NOVA